REAX 1.3
Modified by decisionmodels, 2026: weight-space blending of each part with a Bittensor SN92 competition miner model; averaging of the two parts' per-option probabilities, 50/50. Blend proportions are not disclosed.


=== CharlieBT/gx31-decider @ 161a41004434e7fdad0aa5facb827159ffbe2e4c ===
This model is a modified version of tattrongvu/Jeffan-31b-1710@cebfa86c2bed2b5c83b12e44de4374b20c87fd79 (Apache-2.0), itself derived from google/gemma-4-31B-it. Modified by the submitter of d6s200: full fine-tune of the decoder layers (see MODIFICATIONS.md).


=== vllm-sr/d3-ultra @ 52ba9e7a00402f9560c7fba8c039608e7ddee7fb ===
d3-ultra (Decision 3.0)
Copyright 2026 vLLM Semantic Router Team


=== gurujustin/sn92-gen-p5 @ e2b0728e9699b16e01ae0efeee518ce621696d63 ===
This model is a LoRA fine-tune of ronanlee777/sn92-gen-ma@d4aa443d03e12c073a176c84eb53096898943986 (MIT License), merged
into the backbone weights. That model is itself a 0.5/0.5 weight-space merge of isomsom/sn92-gen-m87b@7f04058f and
xflud/sn92-decider-1@db2e622a, on the common base baselabs/mammoth-1@61cec44c. Its notice, which retains each of its
parents' notices verbatim, is retained verbatim below.

=== ronanlee777/sn92-gen-ma@d4aa443d03e12c073a176c84eb53096898943986 (MIT License) ===
This model is a weight-space merge of the models below; it contains no training of its own.
Each parent's NOTICE is retained verbatim.

=== isomsom/sn92-gen-m87b@7f04058f63745440fae3c10f8da249a0c55fea40 (MIT License), merge weight 0.5 ===
This model is a modified version of works licensed under the MIT License and the Apache License, Version 2.0
(see LICENSE). Its weights were changed from those works.

baselabs/mammoth-1, revision 61cec44c281c693992e3d6ebeb44dd3402dfa186, MIT License (Base Labs, Baseten).
Qwen/Qwen3.5-27B, Apache License 2.0 (the base of mammoth-1).

Modifications: rewritten to the joint readout layout (Qwen3_5Model backbone, the LM head replaced by a readout of the
answer-code rows), LoRA fine-tuned and merged into the backbone weights, readout rows rescaled and shifted.

This model is not affiliated with, endorsed by, or sponsored by Base Labs / Baseten or the Qwen team.

=== xflud/sn92-decider-1@db2e622aa508bd82350620bf165441fa4043f3cf (MIT License), merge weight 0.5 ===
This model is a modified version of works licensed under the MIT License and the Apache License, Version 2.0
(see LICENSE). Its weights were changed from those works.

baselabs/mammoth-1, revision 61cec44c281c693992e3d6ebeb44dd3402dfa186, MIT License.
Qwen/Qwen3.5-27B, Apache License 2.0.
perplexity-ai/pplx-decider-v1.2, revision cce0f3f80033e9d89d80831d5281fd72124aacce, Apache License 2.0
(configuration files). Its notice, retained:
  pplx-decider-v1.2
  Derived from Qwen/Qwen3.8-27B, revision
  1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0, under Apache License 2.0.

This model is not affiliated with, endorsed by, or sponsored by baselabs / Baseten, the Qwen team or Perplexity AI.


=== ronanlee777/sn92-gen-ma @ d4aa443d03e12c073a176c84eb53096898943986 ===
This model is a weight-space merge of the models below; it contains no training of its own.
Each parent's NOTICE is retained verbatim.

=== isomsom/sn92-gen-m87b@7f04058f63745440fae3c10f8da249a0c55fea40 (MIT License), merge weight 0.5 ===
This model is a modified version of works licensed under the MIT License and the Apache License, Version 2.0
(see LICENSE). Its weights were changed from those works.

baselabs/mammoth-1, revision 61cec44c281c693992e3d6ebeb44dd3402dfa186, MIT License (Base Labs, Baseten).
Qwen/Qwen3.5-27B, Apache License 2.0 (the base of mammoth-1).

Modifications: rewritten to the joint readout layout (Qwen3_5Model backbone, the LM head replaced by a readout of the
answer-code rows), LoRA fine-tuned and merged into the backbone weights, readout rows rescaled and shifted.

This model is not affiliated with, endorsed by, or sponsored by Base Labs / Baseten or the Qwen team.

=== xflud/sn92-decider-1@db2e622aa508bd82350620bf165441fa4043f3cf (MIT License), merge weight 0.5 ===
This model is a modified version of works licensed under the MIT License and the Apache License, Version 2.0
(see LICENSE). Its weights were changed from those works.

baselabs/mammoth-1, revision 61cec44c281c693992e3d6ebeb44dd3402dfa186, MIT License.
Qwen/Qwen3.5-27B, Apache License 2.0.
perplexity-ai/pplx-decider-v1.2, revision cce0f3f80033e9d89d80831d5281fd72124aacce, Apache License 2.0
(configuration files). Its notice, retained:
  pplx-decider-v1.2
  Derived from Qwen/Qwen3.8-27B, revision
  1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0, under Apache License 2.0.

This model is not affiliated with, endorsed by, or sponsored by baselabs / Baseten, the Qwen team or Perplexity AI.


=== isomsom/sn92-gen-m87b @ 7f04058f63745440fae3c10f8da249a0c55fea40 ===
This model is a modified version of works licensed under the MIT License and the Apache License, Version 2.0
(see LICENSE). Its weights were changed from those works.

baselabs/mammoth-1, revision 61cec44c281c693992e3d6ebeb44dd3402dfa186, MIT License (Base Labs, Baseten).
Qwen/Qwen3.5-27B, Apache License 2.0 (the base of mammoth-1).

Modifications: rewritten to the joint readout layout (Qwen3_5Model backbone, the LM head replaced by a readout of the
answer-code rows), LoRA fine-tuned and merged into the backbone weights, readout rows rescaled and shifted.

This model is not affiliated with, endorsed by, or sponsored by Base Labs / Baseten or the Qwen team.


=== xflud/sn92-decider-1 @ db2e622aa508bd82350620bf165441fa4043f3cf ===
This model is a modified version of works licensed under the MIT License and the Apache License, Version 2.0
(see LICENSE). Its weights were changed from those works.

baselabs/mammoth-1, revision 61cec44c281c693992e3d6ebeb44dd3402dfa186, MIT License.
Qwen/Qwen3.5-27B, Apache License 2.0.
perplexity-ai/pplx-decider-v1.2, revision cce0f3f80033e9d89d80831d5281fd72124aacce, Apache License 2.0
(configuration files). Its notice, retained:
  pplx-decider-v1.2
  Derived from Qwen/Qwen3.8-27B, revision
  1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0, under Apache License 2.0.

This model is not affiliated with, endorsed by, or sponsored by baselabs / Baseten, the Qwen team or Perplexity AI.


=== perplexity-ai/pplx-decider-v1.2 @ cce0f3f80033e9d89d80831d5281fd72124aacce ===
pplx-decider-v1.2

Derived from Qwen/Qwen3.8-27B, revision
1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0, under Apache License 2.0.